evaluate.py

from tensorflow.keras.datasets import mnist # pyright: ignore[reportMissingImports]
import joblib
from models_helpers import (
    normalize_and_flatten,
    load_model,
    print_metrics,
    plot_confusion,
    subset_classes
)

def evaluate_model(path, X_test, y_test, name, labels):
    model = load_model(path)
    y_pred = model.predict(X_test)
    print(f"\n=== {name} ===")
    print_metrics(y_test, y_pred)
    plot_confusion(y_test, y_pred, labels)

def main():
    # 1. Load test split
    #y = y.ravel()
    
    _,(X_test, y_test) = mnist.load_data()
    X_raw, y     = subset_classes(X_test, y_test, keep_labels=[0,1,2,3,4,5,6,7,8,9],per_class=500)

    # 2. Preprocess 


    X_flat = normalize_and_flatten(X_raw, size=(28,28))
    scaler = load_model("scaler.pkl")
    X_s    = scaler.transform(X_flat)
    labels = ['0','1','2','3','4','5','6','7','8','9']
    # Evaluate
    for fname, title in [
        ("nb_model.pkl",  "Naive Bayes"),
        ("dt_model.pkl",  "Decision Tree"),
        ("mlp_model.pkl", "MLP Classifier")
    ]:
        model = load_model(fname)
        y_pred = model.predict(X_s)
        print(f"\n=== {title} ===")
        print_metrics(y, y_pred)
        plot_confusion(y, y_pred, labels)
if __name__ == "__main__":
    main()